Time correlated single photon counting builds a histogram, not a decay curve. Count rate, repetition rate, the instrument response and excitation density decide whether the lifetime you fit means anything at all.
A lifetime is one of the few numbers in optical characterisation that is close to absolute. It does not depend on how much sample you have, how well the beam is aligned or how efficiently the detector collects, which is why a decay time is a stronger claim than an emission intensity. That independence from the setup disappears the moment the acquisition settings are wrong, and the failure is silent: the fit converges, the residuals look acceptable, and the number is short by a factor you cannot see.
Time correlated single photon counting works by repetition. A pulsed source excites the sample, the electronics measure the delay until a single photon arrives at the detector, and that delay goes into a bin. Repeat several million times and the histogram of arrival times reproduces the decay. Nothing on the screen is a decay curve in the sense of a directly recorded signal; it is a statistical reconstruction, and it is only faithful if the statistics are collected correctly.
Which leads to the first and most common error. If the detected count rate is too high relative to the excitation rate, the timing electronics record the first photon of a pulse period and miss any later ones. Early bins fill preferentially, the histogram is biased towards short delays, and the fitted lifetime comes out too short. The remedy is to attenuate the excitation until the detected rate is a small fraction of the pulse rate, then accumulate for longer. Counting for an hour at the right rate beats counting for a minute at the wrong one.
Three exponentials will fit almost any decay, which is precisely why a three exponential fit proves very little. Start with one component. Add a second only if the residuals show clear structure that the single component cannot follow, and a third only if the same is true again. Look at the weighted residuals along the whole trace rather than at the goodness of fit statistic alone, because a poor fit in the first nanosecond is invisible in a summary number and is often where the physics is.
When a multi exponential fit is genuinely required, be explicit about which average you report. An amplitude weighted average and an intensity weighted average of the same components give different numbers and answer different questions, and papers that omit which one was used cannot be compared with each other.
Finally, check that the sample survived. Run the measurement twice in succession and compare. Photobleaching, oxidation and beam induced degradation all shorten a lifetime progressively, and a decay accumulated over an hour on a sensitive material is an average over a sample that was changing throughout. Where that is a risk, measure in an inert atmosphere or a sealed cell, and reduce the excitation.